Distribution-invariant differential privacy

Differential privacy is becoming one gold standard for protecting the of publicly shared data. It has been widely used in social science, data public health, information technology, and U.S. decennial census. Nevertheless, to guarantee differential privacy, existing methods may unavoidably alter conclusion original analysis, as privatization often changes sample distribution. This phenomenon known trade-off between protection statistical accuracy. In this work, we mitigate by developing a distribution-invariant (DIP) method reconcile both high accuracy strict privacy. As result, any downstream or machine learning task yields essentially same if Numerically, under strictness protection, DIP achieves superior wide range simulation studies real-world benchmarks.